A Novel Air Target Intention Recognition Method Based on Sample Reweighting and Attention-Bi-GRU

Target intent recognition is a vital part of battlefield situational judgment and decision-making. However, existing deep learning-based methods assume that the potential distributions of training and test sets are identical. This assumption overlooks the issue that deviations in test data distribution can lead to a decline in recognition accuracy. To address this problem, this article proposes an intention recognition method for air targets based on bidirectional gated recurrent units (Bi-GRU) and sample reweighting (SR). First, to capture the multidimensional and temporal nature of target attributes along with their varying degrees of influence on intention recognition, a temporal self-attention mechanism is employed to capture time-domain variability. In addition, a Bi-GRU module is used to extract target features. Then, considering the distribution bias of test data, the SR model is applied to eliminate the statistical correlation between relevant and irrelevant features to avoid nonlinear dependence among features. Finally, the effectiveness of the method is demonstrated through ablation and comparison experiments in an air combat scenario. The results clearly indicate that our method outperforms some existing methods.